The Armenian Department of the Calouste Gulbenkian Foundation is pleased to announce that the Western Armenian Treebank was released on 15 May 2021 and is now available within the Universal Dependencies Consortium.
Treebanks are essential tools for ensuring the vitality of a language; they guarantee that the language keeps pace with modern developments in the fields of linguistics and technology. Thanks to treebanks, computer programs are “trained” to recognise the various grammatical and linguistic elements of a sentence — that is, the program recognises the verb, the noun, punctuation marks, and more. Treebanks are repositories of sentences annotated with detailed grammatical notes. Thanks to them, a revolution took place in the 1990s in the international field of Natural Language Processing.
Treebanks play a vital role in modern language processing systems, such as machine translation, semantic analysers, annotations of the component parts of a language, and so on. The Department’s director, Razmik Panossian, explains: “For a language to have spell-checking and grammar-checking tools, or to make machine translation possible, that language must have a treebank. We are glad that the Foundation played a vital role in placing this modern tool at the disposal of those who wish to work in the field of language and technology.”
Universal Dependencies is a programme that develops common cross-linguistic treebanks for different languages, of which both Eastern Armenian and Western Armenian now form part. This programme recognises and annotates the common syntaxes of diverse languages, while also allowing annotations of linguistic particularities.
This new treebank was developed by the ArmTDP team, co-led by Marat Yavrumyan (Yerevan State University) and Hrant Khachatryan (the “YerevaNN” research centre), and is based on the Western Armenian section of the “UD Armenian Syntactic Treebank”. The Western Armenian Treebank joins 202 treebanks in 114 different languages. The treebank has been entirely newly prepared, and can therefore serve as the foundation for any tool created for Western Armenian in the field of Natural Language Processing. An updated and expanded version of the treebank will be released at the end of 2021.
The Western Armenian Treebank currently contains 1,780 sentences and 7.5 million words, based on 110 works by 50 authors, written between 1895 and 2010. Various literary works are included — fiction, correspondence, travel diaries, political and literary lectures and speeches, memoirs, and more. It draws on the repository of the Digital Library of Armenian Literature of the American University of Armenia.
The Western Armenian Treebank, and all the Natural Language Processing tools developed thanks to it, are necessary to equip the language with innovative means and to keep pace with the modern level of technological progress.
For further information about this programme and other initiatives, visit this website https://gulbenkian.pt/armenian-communities and subscribe to our newsletter.
